Real-time Parking Slot Detection for Camera-equipped Vehicles

This paper proposes a novel camera-based approach for parking slot detection with markings, using surround-view cameras mounted on vehicles. By introducing a pipeline of marking-specific image feature extraction and novel filtering stages, we detect parking slot markings in pinhole cameras as well as in cameras with fisheye lenses without using a computationally intensive bird-view transformation. After projecting a compact set of image features into 3D space, our orientation-specific Hough transform finds explicitly the left and right edges of the parking slot markings in desired orientations, which is the basis for marking detection and tracking. We present a novel technique to detect and track the parking slots in the scene in a coherent manner, that preserves the structure of the overall parking layout and its temporal consistency. Our algorithm is designed to run on low cost hardware used in vehicles and it is shown to run at 30fps on ARM CPUs. We validated our algorithm on videos representing real-world scenarios of parking slots including marking occlusion, degradation and different weather conditions.